Comparison Between Some Methods to Select Best Logistic Regression Model (With Application on Heart Disease patients)

Abstract

The study of binary dependent variables have been considered one of important processing because of increasing of phenomena that described with this way . therefore the logistic regression was one of the important methods that represent this type of phenomena. the process of choosing the independent variables which affect binary independent variables is considered very necessary . The study consists of three methods to choose the best logistic regression model , which are the forward method , backward method and the proposed method (factor analysis method) . The goodness of fit of the model after applying each method was tested by using to test, (the deviance and Hossmer – Lemshow ) . The compression of the final results based on three criteria Maximum Likelihood Ratio (MLR) , Akiake information criteria(AIC) and bysian information criteria(BIC) . The final results show that the factors result from the for the proposed method (factor analysis) have an ability to decrease the MLR between than any group of variable was chossed by other methods Consequently , the information criteria which based on MLR such as AIC , BIC given preferences for the factors that result from the proposed method (factor analysis) better than other two methods (forward and backward) .